do not materialize rows of spark DataSet until prompted

[scottdraves]

if a user returns a dataset, current spark integration shows a preview.
this triggers more work than users expect.
instead it should just show the shape of the table (column names and types) and have a button.
if the user clicks, then it collects the rows for a preview.

for example without our magic, spark does this by default with tornadoes_2014.csv:

[Date: string, Time: string ... 9 more fields]

we want that but with all the columns, not "9 more fields".

could do this with a table with 0 rows, and instead of the rows having a button that says to fetch some.
or could show it with text (one column per row), plus the button.

[scottdraves]

If ds is a dataset, then ds.schema.fields has an array of structs with name&type.

[scottdraves]

Generate the preview like this:

def preview(df: org.apache.spark.sql.DataFrame) : TableDisplay = {
    return new TableDisplay(Seq(df.schema.fields.map { col => col.name -> col.dataType.typeName}.toMap))
}

Here's an example:

above the table should have 2 buttons, one that says "show 10 rows" the other says "count rows".
if there is any difficulty, then skip the "count rows" button, we can address in another PR.